Does AI threaten social media's conversational function?

The dominant research framing on AI's effect on social media has been sentiment-and-topic: bots manipulating sentiment, AI amplifying topical coherence, recommender-AI shaping what trends. These framings treat AI as a content-amplification problem — what is being said, how positively or negatively, on what topics. Within those framings the corrective is content moderation, fact-checking, or recommender adjustment.

This misframes the more serious threat. Social media's value is not the content stored on it but its function as a place for conversation, idea-testing, and the surfacing of new positions through interaction. The threat from AI-generated posts is not what they say — it is that they do not constitute moves in a conversation. They are posts shaped like talk that do not participate in talk. The medium continues to host content but progressively loses its function as a medium for talk.

This reframing has practical consequences. Content moderation does not address it, because the AI content can be high-quality, factual, and on-topic and still drain conversationality. Sentiment analysis does not address it, because the AI register is largely neutral. Recommender adjustment does not address it, because engagement metrics reward the AI content that is hollowing out the function. The threat operates at the layer of conversational style, which the available levers do not target.

Habermas's distinction between communicative and strategic action is useful here. Conversational social media at its best is communicative action — oriented toward mutual understanding through reciprocal address. AI-generated posts are not strategic in Habermas's sense either — they are a third category, neither communicative nor strategic, because no agent is addressing anyone with any orientation. They are utterances without the structure of address. Does AI writing lack the internal appeal to attention that humans use? is the same point at the post level.